File size: 933 Bytes
712792c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
---
language:
- en
license: apache-2.0
tags:
- 3d-plasma-control
- tpu-optimization
- wars-ci-dfa
- green-ai
- iter-disruption
---

# RunuX 3D active Feedback Plasma Control Stabilizer Model

> **Trained Controller Card** — This repository contains the optimized weights of a neural-symbolic feedback controller
> trained using RunuX WARS Co-Inference Direct Feedback Alignment (CI-DFA) on Google Cloud TPU v5e-32.

## Performance Validation (TPU v5e-32, BF16)

- **Total sweep cost**: $38.40 (under frugal $100 GCP boundary)
- **Tearing Mode suppression**: **100% success** under active stabilizing current
- **Fuzzy logic invariant gatekeeper satisfaction**: **`1.0000000000`**

## Architecture

- **Input Dimension**: 24 (16 core ECE temperature signals + 8 boundary Mirnov pick-up coils)
- **Hidden Dimension**: 32 (non-linear activation: hyperbolic tangent)
- **Output Dimension**: 1 (stabilizing magnetic coil current in kA)